Data assimilation sensitivity experiments in the East Auckland Current system using 4D-Var

Abstract

This study analyses data assimilative numerical simulations in an eddy-dominated western boundary current: the East Auckland Current (EAuC). The goal is to assess the impact of assimilating surface and subsurface data into a model of the EAuC via running observing system experiments (OSEs). We used the Regional Ocean Modeling System (ROMS) in conjunction with the 4-dimensional variational (4D-Var) data assimilation scheme to incorporate sea surface height (SSH) and temperature (SST), as well as subsurface temperature, salinity and velocity from three moorings located at the upper, mid and lower continental slope using a 7 d assimilation window. Assimilation of surface fields (SSH and SST) reduced SSH root mean square deviation (RMSD) by 25 % in relation to the non-assimilative (NoDA) run. The inclusion of velocity subsurface data further reduced SSH RMSD up- and downstream the moorings by 18 %–25 %. By improving the representation of the mesoscale eddy field, data assimilation increased complex correlation between modelled and observed velocity in all experiments by at least three times. However, the inclusion of temperature and salinity slightly decreased the velocity complex correlation. The assimilative experiments reduced the SST RMSD by 36 % in comparison to the NoDA run. The lack of subsurface temperature for assimilation led to larger RMSD (>1 ∘C) around 100 m in relation to the NoDA run. Comparisons to independent Argo data also showed larger errors at 100 m in experiments that did not assimilate subsurface temperature data. Withholding subsurface temperature forces near-surface average negative temperature increments to the initial conditions that are corrected by increased net heat flux at the surface, but this had limited or no effect on water temperature at 100 m depth. Assimilation of mooring temperature generates mean positive increments to the initial conditions that reduces 100 m water temperature RMSD. In addition, negative heat flux and positive wind stress curl were generated near the moorings in experiments that assimilated subsurface temperature data. Positive wind stress curl generates convergence and downwelling that can correct interior temperature but might also be responsible for decreased velocity correlations. The few moored CTDs (eight) had little impact in correcting salinity in comparison to independent Argo data. However, using doubled decorrelation length scales of tracers and a 2 d assimilation window improved model salinity and temperature in comparison to Argo profiles throughout the domain. This assimilation configuration, however, led to large errors when subsurface temperature data were not assimilated due to incorrect increments to the subsurface. As all reanalyses show improved model-observation skill relative to HYCOM–NCODA (the model boundary conditions), these results highlight the benefit of numerical downscaling to a regional model of the EAuC.